Open Source Can't Be Stopped | Charlie O'Neill Explains
Jul 28, 2026 · 14:23
Charlie O'Neill of Baseten argues that open and closed source AI labs can coexist, with Kimi K3 proving open source models can approach frontier capability. He explains the Nvidia open letter's significance as a landmark moment acknowledging this coexistence. O'Neill notes that Kimi K3's training on persistent agent environments—coding, terminals, simulated workplace tools—marks a shift from training on answers to training on work. He claims post-training becomes counterintuitively cheaper when starting from a frontier baseline, as less effort is needed to fix general deficiencies. He predicts 3 to 5 new 2 to 4 trillion-parameter open source models from American labs in the next six months, all engineered for long-running tasks.
